Key Default Type Description
compaction.bucket-distribution-strategy
 linear

Enum

 Defines how dedicated bucket compaction jobs distribute compact buckets to writers. 'linear' uses the existing stable partition-plus-bucket mapping. 'size-aware-batch' assigns bounded full-compaction bucket splits by total data file size and forwards them to writers to reduce compaction long tail.

Possible values:
  • "linear": Distribute compact buckets by the existing stable partition-plus-bucket channel mapping.
  • "size-aware-batch": For bounded full compaction, assign compact bucket splits by total data file size and forward them to writers to reduce long-tail compaction tasks.
changelog.precommit-compact.thread-num
 (none) Integer Maximum number of threads to copy bytes from small changelog files. By default is the number of processors available to the Java virtual machine.
commit.custom-listeners
 (none) String Commit listener will be called after a successful commit. This option list custom commit listener identifiers separated by comma.
end-input.watermark
 (none) Long Optional endInput watermark used in case of batch mode or bounded stream.
filesystem.job-level-settings.enabled
 true Boolean Enable pass job level filesystem settings to table file IO.
lookup.async
 false Boolean Whether to enable async lookup join.
lookup.async-thread-number
 16 Integer The thread number for lookup async.
lookup.bootstrap-parallelism
 4 Integer The parallelism for bootstrap in a single task for lookup join.
lookup.cache
 AUTO

Enum

 The cache mode of lookup join.

Possible values:
  • "AUTO"
  • "FULL"
  • "MEMORY"
lookup.dynamic-partition.refresh-interval
 1 h Duration Specific dynamic partition refresh interval for lookup, scan all partitions and obtain corresponding partition.
lookup.dynamic-partition.refresh.async
 false Boolean Whether to refresh dynamic partition lookup table asynchronously. This option only works for full cache dimension table. When enabled, partition changes will be loaded in a background thread while the old partition data continues serving queries. When disabled (default), partition refresh is synchronous and blocks queries until the new partition data is fully loaded.
lookup.refresh.async
 false Boolean Whether to refresh lookup table in an async thread.
lookup.refresh.async.pending-snapshot-count
 5 Integer If the pending snapshot count exceeds the threshold, lookup operator will refresh the table in sync.
lookup.refresh.full-load-threshold
 (none) Integer If the pending snapshot count exceeds this threshold, lookup table will discard incremental updates and refresh the entire table from the latest snapshot. This can improve performance when there are many snapshots pending. Set to a reasonable value (e.g., 10) to enable this optimization. Default is Integer.MAX_VALUE (disabled).
lookup.refresh.time-periods-blacklist
 (none) String The blacklist contains several time periods. During these time periods, the lookup table's cache refreshing is forbidden. Blacklist format is start1->end1,start2->end2,... , and the time format is yyyy-MM-dd HH:mm. Only used when lookup table is FULL cache mode.
partition.idle-time-to-done
 (none) Duration Set a time duration when a partition has no new data after this time duration, mark the done status to indicate that the data is ready.
partition.mark-done-action.mode
 process-time

Enum

 How to trigger partition mark done action.

Possible values:
  • "process-time": Based on the time of the machine, mark the partition done once the processing time passes period time plus delay.
  • "watermark": Based on the watermark of the input, mark the partition done once the watermark passes period time plus delay.
partition.mark-done.recover-from-state
 true Boolean Whether trigger partition mark done when recover from state.
partition.time-interval
 (none) Duration You can specify time interval for partition, for example, daily partition is '1 d', hourly partition is '1 h'.
precommit-compact
 false Boolean If true, it will add a compact coordinator and worker operator after the writer operator,in order to compact several changelog files (for primary key tables) or newly created data files (for unaware bucket tables) from the same partition into large ones, which can decrease the number of small files.
read.shuffle-bucket-with-partition
 true Boolean Whether shuffle by partition and bucket when read.
scan.bounded
 (none) Boolean Bounded mode for Paimon consumer. By default, Paimon automatically selects bounded mode based on the mode of the Flink job.
scan.dedicated-split-generation
 false Boolean If true, the split generation process would be performed during runtime on a Flink task, instead of on the JobManager during initialization phase.
scan.infer-parallelism
 true Boolean If it is false, parallelism of source are set by global parallelism. Otherwise, source parallelism is inferred from splits number (batch mode) or bucket number(streaming mode).
scan.infer-parallelism.max
 1024 Integer If scan.infer-parallelism is true, limit the parallelism of source through this option.
scan.max-snapshot.count
 -1 Integer The max snapshot count to scan per checkpoint. Not limited when it's negative.
scan.parallelism
 (none) Integer Define a custom parallelism for the scan source. By default, if this option is not defined, the planner will derive the parallelism for each statement individually by also considering the global configuration. If user enable the scan.infer-parallelism, the planner will derive the parallelism by inferred parallelism.
scan.partitions
 (none) String Specify the partitions to scan. Partitions should be given in the form of key1=value1,key2=value2. Partition keys not specified will be filled with the value of partition.default-name. Multiple partitions should be separated by semicolon (;). This option can support normal source tables and lookup join tables. There are two special values max_pt() and max_two_pt() are also supported to specify the (two) partition(s) with the largest partition value. For lookup source, the max partition(s) will be periodically refreshed; for normal source, the max partition(s) will be determined before job running without refreshing even for streaming jobs.
scan.remove-normalize
 false Boolean Whether to force the removal of the normalize node when streaming read. Note: This is dangerous and is likely to cause data errors if downstream is used to calculate aggregation and the input is not complete changelog.
scan.split-enumerator.batch-size
 10 Integer How many splits should assign to subtask per batch in StaticFileStoreSplitEnumerator to avoid exceed `akka.framesize` limit.
scan.split-enumerator.mode
 fair

Enum

 The mode used by StaticFileStoreSplitEnumerator to assign splits.

Possible values:
  • "fair": Distribute splits evenly when batch reading to prevent a few tasks from reading all.
  • "preemptive": Distribute splits preemptively according to the consumption speed of the task.
scan.watermark.alignment.group
 (none) String A group of sources to align watermarks.
scan.watermark.alignment.max-drift
 (none) Duration Maximal drift to align watermarks, before we pause consuming from the source/task/partition.
scan.watermark.alignment.update-interval
 1 s Duration How often tasks should notify coordinator about the current watermark and how often the coordinator should announce the maximal aligned watermark.
scan.watermark.emit.strategy
 on-event

Enum

 Emit strategy for watermark generation.

Possible values:
  • "on-periodic": Emit watermark periodically, interval is controlled by Flink 'pipeline.auto-watermark-interval'.
  • "on-event": Emit watermark per record.
scan.watermark.idle-timeout
 (none) Duration If no records flow in a partition of a stream for that amount of time, then that partition is considered "idle" and will not hold back the progress of watermarks in downstream operators.
sink.clustering.sample-factor
 100 Integer Specifies the sample factor. Let S represent the total number of samples, F represent the sample factor, and P represent the sink parallelism, then S=F×P. The minimum allowed sample factor is 20.
sink.clustering.sort-in-cluster
 true Boolean Indicates whether to further sort data belonged to each sink task after range partitioning.
sink.committer-cpu
 1.0 Double Sink committer cpu to control cpu cores of global committer.
sink.committer-memory
 (none) MemorySize Sink committer memory to control heap memory of global committer.
sink.committer-operator-chaining
 true Boolean Allow sink committer and writer operator to be chained together
sink.coordinator-commit.enabled
 false Boolean If true, run the Paimon committer inside the Flink JobManager via an OperatorCoordinator. This decouples commit from any single TaskManager subtask so that region failover does not have to restart the whole pipeline. Only supports unaware-bucket append tables in streaming mode with checkpointing enabled; unsupported configurations fail during sink planning.
sink.cross-partition.managed-memory
 256 mb MemorySize Weight of managed memory for the local key-value index in cross-partition update, Flink will compute the memory size according to the weight, the actual memory used depends on the running environment.
sink.key-only-deletes.enabled
 false Boolean If true, a primary-key table sink advertises the key-only (partial) deletes capability, allowing the Flink planner to drop the upstream ChangelogNormalize node when the source produces deletes by key. Requires Flink 2.1+; no effect on Flink 1.x or 2.0. Does not apply when the table has no primary key, when 'changelog-producer' is 'input', or when 'merge-engine' is 'aggregation' or 'partial-update' with aggregation functions; in those cases a warning is logged. Disabled by default.
sink.managed.writer-buffer-memory
 256 mb MemorySize Weight of writer buffer in managed memory, Flink will compute the memory size for writer according to the weight, the actual memory used depends on the running environment.
sink.operator-uid.suffix
 (none) String Set the uid suffix for the writer, dynamic bucket assigner and committer operators. The uid format is ${UID_PREFIX}_${TABLE_NAME}_${USER_UID_SUFFIX}. If the uid suffix is not set, flink will automatically generate the operator uid, which may be incompatible when the topology changes.
sink.parallelism
 (none) Integer Defines a custom parallelism for the sink. By default, if this option is not defined, the planner will derive the parallelism for each statement individually by also considering the global configuration.
sink.savepoint.auto-tag
 false Boolean If true, a tag will be automatically created for the snapshot created by flink savepoint.
sink.use-managed-memory-allocator
 false Boolean If true, flink sink will use managed memory for merge tree; otherwise, it will create an independent memory allocator.
sink.writer-coordinator.cache-expire-after-access
 (none) Duration Optional idle TTL for writer coordinator manifest cache entries. Disabled by default. When set, an entry that has not been accessed within this duration is evicted, releasing its heap. The cache stays bounded by 'sink.writer-coordinator.cache-memory' regardless of this setting.
sink.writer-coordinator.cache-memory
 2 gb MemorySize Controls the cache memory of writer coordinator to cache manifest files in Job Manager.
sink.writer-coordinator.cache-soft-values
 true Boolean If true (default), writer coordinator manifest cache entries are held with soft references and may be reclaimed by the GC under memory pressure. This can trigger a cache-thrash spiral where reclaimed entries are refetched, spiking heap and forcing further reclamation. Set to false to hold entries with strong references, breaking the spiral; the cache then stays bounded by weight up to 'sink.writer-coordinator.cache-memory' (size the Job Manager total heap memory to at least roughly twice that value).
sink.writer-coordinator.enabled
 false Boolean Enable sink writer coordinator to plan data files in Job Manager.
sink.writer-coordinator.page-size
 32 kb MemorySize Controls the page size for one RPC request of writer coordinator.
sink.writer-coordinator.prefetch-manifests
 false Boolean If true, the writer coordinator eagerly reads all data manifests of the latest snapshot during refresh to warm the in-Job-Manager manifest cache. This avoids many concurrent cold manifest reads when high-parallelism writers restore at the same time, reducing Job Manager heap pressure at the cost of one full manifest read per refresh.
sink.writer-cpu
 1.0 Double Sink writer cpu to control cpu cores of writer.
sink.writer-memory
 (none) MemorySize Sink writer memory to control heap memory of writer.
sink.writer-refresh-detectors
 (none) String The option groups which are expected to be refreshed when streaming writing, multiple option group separated by commas. Now only 'external-paths' is supported.
source.checkpoint-align.enabled
 false Boolean Whether to align the flink checkpoint with the snapshot of the paimon table, If true, a checkpoint will only be made if a snapshot is consumed.
source.checkpoint-align.timeout
 30 s Duration If the new snapshot has not been generated when the checkpoint starts to trigger, the enumerator will block the checkpoint and wait for the new snapshot. Set the maximum waiting time to avoid infinite waiting, if timeout, the checkpoint will fail. Note that it should be set smaller than the checkpoint timeout.
source.operator-uid.suffix
 (none) String Set the uid suffix for the source operators. After setting, the uid format is ${UID_PREFIX}_${TABLE_NAME}_${USER_UID_SUFFIX}. If the uid suffix is not set, flink will automatically generate the operator uid, which may be incompatible when the topology changes.
unaware-bucket.compaction.parallelism
 (none) Integer Defines a custom parallelism for the unaware-bucket table compaction job. By default, if this option is not defined, the planner will derive the parallelism for each statement individually by also considering the global configuration.
unaware-bucket.no-shuffle
 false Boolean If true, the CDC sync pipeline will skip the network shuffle between source and writer operators. This is only supported for bucket-unaware (append) tables where each writer subtask independently appends data without bucket ownership constraints. This eliminates data transfer overhead when the source already provides suitable data distribution (e.g., Kafka partitions).